Abstract
This paper presents a novel Hybrid Deep Neural Network approach for enhancing face liveness detection in video sequences, a critical component in safeguarding biometric authentication systems against sophisticated spoofing attacks. Leveraging the strengths of convolutional neural networks for spatial and frequency domain feature extraction the proposed model offers a comprehensive solution to detect and differentiate between live and spoofed face presentations. Utilizing the OULU-NPU dataset, featuring real-world variations and diverse spoofing scenarios, our model was rigorously evaluated across multiple performance metrics, including accuracy, precision, recall, and F1-score. The research introduces dynamic thresholding and Grad-CAM visualizations to improve decision-making flexibility and model interpretability, respectively. The results demonstrate the HDNN model's significant performance and adaptability to various attack vectors and environmental conditions, making a significant contribution to the field of biometric security. In addition to advancing face liveness detection technologies, this work paves the way for future investigations into biometric identification systems that are more reliable and effective.